I was a little disappointed that they didn't show any true video understanding to differentiate between the model being capable of taking photos when instructed or actually having a frame-by-frame understanding of what's happening.

If I were to roll a ball across the screen would it know the direction?